AI Operations Engineer

CR Fence & Rail; India; A distancia; INR 900.000 - 1.500.000; Jornada completa

Descripción de la vacante

CR Fence & Rail is looking for an AI-Driven Platform Owner to lead our ticketing and customer-support AI platform end-to-end. You’ll influence architecture, train and tune AI models, and ensure reliability across email, chat, voice and SMS channels.

You’ll rely on agentic AI tools to craft and verify changes, read outputs, and iterate with human review. The role emphasizes systems thinking, operational understanding, and business impact over handwritten code.

Formación

• Must be fluent with agentic AI coding tools (e.g., Claude Code) and able to guide AI to produce correct results.
• Strong technical judgment: read code and understand system fit, detect unsafe or wrong changes.
• Verification mindset: test what you ship and ensure it works in real scenarios.
• Clear written communication: describe problems and outcomes precisely.
• Solid software fundamentals: understand web apps, databases, and APIs.

Responsabilidades

• Multi-channel ingestion & threading to pull messages from Gmail, SMTP/IMAP, Twilio, and web chat.
• The AI customer-support layer: intent classification, response drafting, guardrails, and calibration loops.
• Assignment, routing & SLA automation with auto-assign and specialist routing rules.
• CSAT: manage post-resolution surveys and AI analysis of feedback.
• Voice subsystem: real-time Twilio integration, voicemail transcription, and ticket linkage.
• React frontend: agent-facing surface including ticket detail, composer, inboxes, queues, dashboards.

Conocimientos

Agentic AI tools

Technical judgment

Verification mindset

Clear communication

Software fundamentals

Herramientas

TypeScript/Node

React

PostgreSQL/Supabase

AI/ML basics

Descripción del empleo

How we work

Team: Operations Engineering ·

Type: Full-time ·

Location: Remote

Reports to: Lead Developer (partners closely with the Head of Operations)

Owns: The Ticketing & Customer-Support AI platform

We run a high-volume, AI-powered customer operations platform built entirely in-house — a full Zendesk/Gorgias-class ticketing system with an AI customer-support agent layered on top. It handles conversations across email, SMS, voice, WhatsApp and live chat, threads and routes them, drafts AI responses for human approval, and measures satisfaction afterwards.

Today one person covers it. We're hiring the person who will own it end to end.

Here's the important part: we build with AI, not by hand. This is an AI-assisted development role. You'll direct agentic AI coding tools to make changes, then read, test and verify what they produce. The day-to-day is far more about clear thinking, good instructions and careful review than about typing code yourself. If you can describe exactly what needs to happen, spot when the AI got it subtly wrong, and keep a live system healthy, this role is for you — regardless of how many lines of code you personally write.

This is not a bounded \"work a ticket queue\" job. You will own one of the largest and most business-critical systems we run, and you'll be measured on business outcomes, not commits.

You'll take over from our current owner through a structured handoff: routing rules and SLA triggers, the AI classification model and its training data, and a minimum two-week pair-working ramp. The mandate afterwards is clear — keep the platform reliable, and push more of the support workload onto AI.

• AI writes the code; you drive it. We use agentic AI development tools (Claude Code and similar). Your leverage comes from breaking down what needs to happen, prompting well, and reviewing the result — not from hand-writing the implementation.


• Ownership over implementation. We value systems thinking, operational understanding and business impact far more than lines of code produced.


• Verify everything. Nothing is \"done\" until it has been tested and shown working against real behaviour, before it reaches production. A big part of this job is catching the cases where AI output looks right but isn't.


• Understand before you change. You'll spend real time reading how the system works so your instructions land in the right place. Judgment about what to change matters more than raw coding speed.

What you’ll own

• Multi-channel ingestion & threading — the engine that pulls messages from Gmail (OAuth + push), SMTP/IMAP, Twilio (SMS, voice, WhatsApp) and web chat, then threads them into conversations using message headers, customer matching and subject similarity, with deduplication and reopen logic.


• The AI customer-support layer — intent classification, response drafting, guardrails, and a calibration loop that learns from the edits human reps make to AI drafts. The goal: AI drafts a response, a rep approves it in under two minutes.


• Assignment, routing & SLA automation — presence-aware auto-assignment across the CS team, round-robin and specialist routing rules, idle-unassign, reservation logic, and the SLA triggers that keep work moving.


• CSAT — post-resolution satisfaction surveys: timezone- and business-hours-aware delivery, reply parsing, and AI analysis of the feedback.


• The voice subsystem — a real-time Twilio Media Streams + TaskRouter pipeline, including AI that answers calls from our knowledge base, voicemail transcription, and call-to-ticket linkage.


• The React frontend — the agent-facing surface reps live in all day: ticket detail, message composer, inboxes, queues, dashboards and reporting.

You’ll also work across the neighbouring AI modules this platform depends on — the customer-support orchestrator, AI guidance, and knowledge‑base agents — and against the internal systems it reads from (orders, fulfilment and shipping). Comfort reading into adjacent systems is part of the job.

What we're looking for

Must have

• Fluency with agentic AI coding tools (Claude Code, Cursor or similar) — you've used them to build or change real software and you know how to get good results from them. (Autocomplete-style tooling alone isn't the same thing.)


• Strong technical judgment — you can read code and understand how a system fits together well enough to know what to change, even when the AI writes the change. You can tell when generated code is wrong, unsafe, or a band‑aid.


• A verification mindset — you test what you ship and don't call something done until you've seen it work.


• Clear written communication — the quality of your instructions is the quality of the output. You can describe a problem and the desired outcome precisely.


• Enough software fundamentals to be dangerous — comfortable with how web apps, databases and APIs work, and able to follow data through a system.

Strongly preferred

• Familiarity with our kind of stack — TypeScript/Node, React, PostgreSQL/Supabase — enough to review changes confidently.


• Experience with LLM/AI features (we use Anthropic Claude): prompts, classification, guardrails, and judging output quality.


• Exposure to third-party integrations — email, telephony (Twilio), messaging, REST APIs, webhooks, OAuth — and the ways they fail.


• Background in customer-support / helpdesk / CRM / ecommerce software, or any high-volume messaging system.

You do not need to be a traditional \"hardcore\" programmer, to have memorised a framework, or to write large amounts of code from scratch. The AI does the typing. We're hiring for ownership, analytical thinking, AI collaboration, and the ability to keep improving a complex operational system.

The stack the AI works in: TypeScript · Node.js / Express · React · PostgreSQL / Supabase · Anthropic Claude · Twilio · Gmail & SMTP/IMAP · Playwright · Docker · DigitalOcean

Success in this role

Grounded in the targets we already track for this seat:

• Ownership & handoff — within the first weeks, absorb the full platform via the documented handoff and pair‑coding ramp, and become the reliable point of contact for anything ticketing-related.


• AI-handled tickets 40% by end of 2026 — expand the AI draft‑and‑approve flow so it handles a growing share of volume, at a sub-2-minute rep approval time.


• Self-service deflection 20% by end of 2026 — grow the share of customer issues resolved via FAQ / knowledge base / self‑service before a ticket is ever created.


• Keep the platform inside CS SLAs — reliably support first response under 4 business hours and resolution under 24 hours for standard issues.

Why join us

We're building an AI-first ecommerce company where automation replaces repetitive work and engineers own business outcomes, not just code. You'll have a system of real consequence to yourself, modern AI tooling to move it fast, and a direct line between what you ship and what customers experience.